Function _evaluate_schema_driven

python/agents/invoice-processing/eval/eval.py:708–739

Evaluate a case using schema-driven comparison groups from master data.

(
    case_id: str,
    gt: dict,
    agent: dict,
    decision: dict,
    master: MasterData,
    tolerance: float,
)

def _evaluate_schema_driven(
    case_id: str,
    gt: dict,
    agent: dict,
    decision: dict,
    master: MasterData,
    tolerance: float,
) -> dict:
    """Evaluate a case using schema-driven comparison groups from master data."""
    group_results = _run_comparison_groups(gt, agent, master, tolerance)
    mismatches = _collect_schema_mismatches(decision, group_results)

    all_correct = decision["match"] and all(
        r.get("all_match", True) for r in group_results.values()
    )

    result_dict = {
        "case_id": case_id,
        "status": "PASS" if all_correct else "FAIL",
        "all_correct": all_correct,
        "decision": decision,
        "mismatches": mismatches,
    }
    for gid, result in group_results.items():
        result_dict[gid] = result

    # Keep backward-compatible keys for the standard invoice groups
    for compat_key in ("financials", "vendor", "header", "line_items"):
        if compat_key in group_results:
            result_dict[compat_key] = group_results[compat_key]

    return result_dict


def _evaluate_legacy(